How it works?
How the Eave Engine Works
So how does all this magic actually happen? Think of the Eave Engine as a multi-stage factory, where raw, messy web data comes in one side — and clean and structured data come out the other.
Here’s a breakdown of the key stages (no buzzwords, just facts):
Transcription & Sentence Splitting: Turning Speech into Text
First, we capture real conversations — But instead of handing you one giant wall of text, we split these conversations into smart, manageable chunks.
Why? Because AI models (and humans) hate reading endless paragraphs.
Every piece gets capped at 1,500 tokens, so we keep the context tight and sharp — no loss of meaning.
Result: Structured, digestible slices of conversations.
Autocorrection: Fixing the Mess Humans Make
People don’t talk like books. They ramble, misspeak, and repeat themselves. Our autocorrection layer runs on a fine-tuned Large Language Model (LLM) that:
Fixes grammar and spelling.
Normalizes speaker names (so "Elon" and "Elon Musk" don’t show up as two different people).
Standardizes crypto slang and jargon (because yes, we know what "rekt" means).
Result: Clean, readable, and context-accurate transcripts — without losing the speaker’s original tone.
Semantic Chunking: Keeping the Story Together
Once everything is corrected, we semantically chunk the text — meaning we break it up by meaning, not just random sentences.
Embeddings (via OpenAI’s
text-embedding-3-small) help us find where conversations naturally shift.No more cutting topics in half — we preserve the flow of the conversation.
Result: Data that makes sense when you read it — as if you were there in real-time.
Entity Recognition: Who’s in the Room? What’s Being Talked About?
We scan every chunk for key entities — from "Bitcoin" to "Solana" to lesser-known projects.
We pull tickers, project names, speakers, companies, and more.
Not just what's said, but who's saying it — crucial for understanding influence and sentiment.
Result: Structured lists of topics, people, and projects — fully tagged and searchable.
Sentiment & Signal Detection: Is the Room Bullish or Bearish?
With entities in hand, we run Crypto-specific sentiment analysis to detect:
Bullish / Bearish / Neutral tones.
Trading signals like buy, sell, hold, stop-loss.
And even scam warnings and market alerts buried in conversations.
Result: Real-time insights into what the market is thinking — before it moves.
GraphRAG & Entity Linking: Making Sense of It All
Finally, we stitch everything together in a graph-based index (GraphRAG):
Entities are linked to external sources (market caps, official sites, tickers).
We map relationships between projects, people, and narratives — so you don’t just know what was said, but how it connects.
Result: A navigable, queryable map of the Web3 world.
In short? The Eave Engine doesn’t just collect data — it understands it
Last updated